// setKeys sets n random keys and values across each machine in a
// cluster and returns these values to later be checked with checkKeys.
// If all the values don't get set due to a machine that is down and
// error is NOT returned. An error is returned if no keys are able to be
// set.
func SetKeys(cluster platform.Cluster, n int) (map[string]string, error) {
var written = map[string]string{}
for _, m := range cluster.Machines() {
for i := 0; i < n; i++ {
// random key and value, may overwrwite previous sets if
// collision which is fine
key := strconv.Itoa(rand.Int())[0:3]
value := strconv.Itoa(rand.Int())[0:3]
b, err := m.SSH(fmt.Sprintf("curl -s -w %%{http_code} -s http://127.0.0.1:2379/v2/keys/%v -XPUT -d value=%v", key, value))
if err != nil {
return nil, err
}
// check for 201 or 200 resp header
if !bytes.HasSuffix(b, []byte("200")) && !bytes.HasSuffix(b, []byte("201")) {
continue
}
written[key] = value
}
}
if len(written) == 0 {
return nil, fmt.Errorf("failed to write any keys")
}
plog.Infof("wrote %v keys", len(written))
return written, nil
}
// setKeys sets n random keys and values across each machine in a
// cluster and returns these values to later be checked with checkKeys.
// If all the values don't get set due to a machine that is down and
// error is NOT returned. An error is returned if no keys are able to be
// set.
func setKeys(cluster platform.Cluster, n int) (map[string]string, error) {
var written = map[string]string{}
for _, m := range cluster.Machines() {
for i := 0; i < n; i++ {
// random key and value, may overwrwite previous sets if
// collision which is fine
key := strconv.Itoa(rand.Int())[0:3]
value := strconv.Itoa(rand.Int())[0:3]
cmd := cluster.NewCommand("curl", "-w", "%{http_code}", "-s", fmt.Sprintf("http://%v:2379/v2/keys/%v", m.IP(), key), "-XPUT", "-d", "value="+value)
b, err := cmd.Output()
if err != nil {
continue
}
// check for 201 or 200 resp header
if !bytes.HasSuffix(b, []byte("200")) && !bytes.HasSuffix(b, []byte("201")) {
continue
}
written[key] = value
}
}
if len(written) == 0 {
return nil, fmt.Errorf("failed to write any keys")
}
plog.Infof("wrote %v keys", len(written))
return written, nil
}
// checkKeys tests that each node in the cluster has the full provided
// key set in keyMap. Quorum get must be used.
func CheckKeys(cluster platform.Cluster, keyMap map[string]string, quorum bool) error {
for i, m := range cluster.Machines() {
for k, v := range keyMap {
var cmd string
if quorum {
cmd = fmt.Sprintf("curl http://127.0.0.1:2379/v2/keys/%v?quorum=true", k)
} else {
cmd = fmt.Sprintf("curl http://127.0.0.1:2379/v2/keys/%v", k)
}
s, err := m.SSHSession()
if err != nil {
return err
}
defer s.Close()
b, err := s.Output(cmd)
if err != nil {
return fmt.Errorf("error curling key: %v", err)
}
var jsonMap map[string]interface{}
err = json.Unmarshal(b, &jsonMap)
if err != nil {
return err
}
// error code?
errorCode, ok := jsonMap["errorCode"]
if ok {
msg := jsonMap["message"]
return fmt.Errorf("machine %v errorCode %v: %v: %s", i, errorCode, msg, b)
}
node, ok := jsonMap["node"]
if !ok {
return fmt.Errorf("retrieving key in CheckKeys, no node in resp")
}
n := node.(map[string]interface{})
value, ok := n["value"]
if !ok {
return fmt.Errorf("retrieving key in CheckKeys, no value in resp")
}
if value != v {
return fmt.Errorf("checkKeys got incorrect value! expected:%v got: %v", v, value)
}
}
}
plog.Infof("checked %v keys", len(keyMap))
return nil
}
// checkKeys tests that each node in the cluster has the full provided
// key set in keyMap. Quorum get must be used.
func checkKeys(cluster platform.Cluster, keyMap map[string]string) error {
for i, m := range cluster.Machines() {
for k, v := range keyMap {
cmd := cluster.NewCommand("curl", fmt.Sprintf("http://%v:2379/v2/keys/%v?quorum=true", m.IP(), k))
b, err := cmd.Output()
if err != nil {
return fmt.Errorf("error curling key: %v", err)
}
var jsonMap map[string]interface{}
err = json.Unmarshal(b, &jsonMap)
if err != nil {
return err
}
// error code?
errorCode, ok := jsonMap["errorCode"]
if ok {
msg := jsonMap["message"]
return fmt.Errorf("machine %v errorCode %v: %v: %s", i, errorCode, msg, b)
}
node, ok := jsonMap["node"]
if !ok {
return fmt.Errorf("retrieving key in CheckKeys, no node in resp")
}
n := node.(map[string]interface{})
value, ok := n["value"]
if !ok {
return fmt.Errorf("retrieving key in CheckKeys, no value in resp")
}
if value != v {
return fmt.Errorf("checkKeys got incorrect value! expected:%v got: %v", v, value)
}
}
}
plog.Infof("checked %v keys", len(keyMap))
return nil
}
func discovery(cluster platform.Cluster, version int) error {
if plog.LevelAt(capnslog.DEBUG) {
// get journalctl -f from all machines before starting
for _, m := range cluster.Machines() {
if err := platform.StreamJournal(m); err != nil {
return fmt.Errorf("failed to start journal: %v", err)
}
}
}
// start etcd on each machine asynchronously.
for _, m := range cluster.Machines() {
if err := doStart(m, version, false); err != nil {
return err
}
}
// block until each instance is reported as started.
for i, m := range cluster.Machines() {
if err := doStart(m, version, true); err != nil {
return err
}
plog.Infof("etcd instance%d started", i)
}
var keyMap map[string]string
var retryFuncs []func() error
retryFuncs = append(retryFuncs, func() error {
var err error
keyMap, err = SetKeys(cluster, 5)
if err != nil {
return err
}
return nil
})
retryFuncs = append(retryFuncs, func() error {
var quorumRead bool
if version == 2 {
quorumRead = true
}
if err := CheckKeys(cluster, keyMap, quorumRead); err != nil {
return err
}
return nil
})
for _, retry := range retryFuncs {
if err := util.Retry(5, 5*time.Second, retry); err != nil {
return fmt.Errorf("discovery failed health check: %v", err)
}
// NOTE(pb): etcd1 seems to fail in an odd way when I try quorum
// read, instead just sleep between setting and getting.
time.Sleep(2 * time.Second)
}
return nil
}
func discovery(cluster platform.Cluster, version int) error {
csize := len(cluster.Machines())
if plog.LevelAt(capnslog.DEBUG) {
// get journalctl -f from all machines before starting
for _, m := range cluster.Machines() {
if err := m.StartJournal(); err != nil {
return fmt.Errorf("failed to start journal: %v", err)
}
}
}
// point etcd on each machine to discovery
for i, m := range cluster.Machines() {
// start etcd instance
var etcdStart string
if version == 1 {
etcdStart = "sudo systemctl start etcd.service"
} else if version == 2 {
etcdStart = "sudo systemctl start etcd2.service"
} else {
return fmt.Errorf("etcd version unspecified")
}
_, err := m.SSH(etcdStart)
if err != nil {
return fmt.Errorf("SSH cmd to %v failed: %s", m.IP(), err)
}
plog.Infof("etcd instance%d started", i)
}
err := getClusterHealth(cluster.Machines()[0], csize)
if err != nil {
return fmt.Errorf("discovery failed health check: %v", err)
}
return nil
}